Adversarial Robustness on Surface-form evasion probe suite Mixed-language mentions
38.8Exposure RateRegex-only extractor
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Regex-only extractorType=Baseline2026.04 | 38.8 | — | |
| BodhiPromptShieldPolicy Mediator=Normalization-aware2026.04 | 38.8 | 58.3 |